arch/x86_64/kernel/traps.c

   1 /*
   2  *  linux/arch/x86-64/traps.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   6  *
   7  *  Pentium III FXSR, SSE support
   8  *      Gareth Hughes <gareth@valinux.com>, May 2000
   9  *
  10  *  $Id: traps.c,v 1.36 2002/03/24 11:09:10 ak Exp $
  11  */
  12
  13 /*
  14  * 'Traps.c' handles hardware traps and faults after we have saved some
  15  * state in 'entry.S'.
  16  */
  17 #include <linux/config.h>
  18 #include <linux/sched.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/timer.h>
  24 #include <linux/mm.h>
  25 #include <linux/init.h>
  26 #include <linux/delay.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/module.h>
  30 #include <linux/moduleparam.h>
  31
  32 #include <asm/system.h>
  33 #include <asm/uaccess.h>
  34 #include <asm/io.h>
  35 #include <asm/atomic.h>
  36 #include <asm/debugreg.h>
  37 #include <asm/desc.h>
  38 #include <asm/i387.h>
  39 #include <asm/kdebug.h>
  40 #include <asm/processor.h>
  41
  42 #include <asm/smp.h>
  43 #include <asm/pgalloc.h>
  44 #include <asm/pda.h>
  45 #include <asm/proto.h>
  46 #include <asm/nmi.h>
  47
  48 #include <linux/irq.h>
  49
  50
  51 extern struct gate_struct idt_table[256];
  52
  53 asmlinkage void divide_error(void);
  54 asmlinkage void debug(void);
  55 asmlinkage void nmi(void);
  56 asmlinkage void int3(void);
  57 asmlinkage void overflow(void);
  58 asmlinkage void bounds(void);
  59 asmlinkage void invalid_op(void);
  60 asmlinkage void device_not_available(void);
  61 asmlinkage void double_fault(void);
  62 asmlinkage void coprocessor_segment_overrun(void);
  63 asmlinkage void invalid_TSS(void);
  64 asmlinkage void segment_not_present(void);
  65 asmlinkage void stack_segment(void);
  66 asmlinkage void general_protection(void);
  67 asmlinkage void page_fault(void);
  68 asmlinkage void coprocessor_error(void);
  69 asmlinkage void simd_coprocessor_error(void);
  70 asmlinkage void reserved(void);
  71 asmlinkage void alignment_check(void);
  72 asmlinkage void machine_check(void);
  73 asmlinkage void spurious_interrupt_bug(void);
  74 asmlinkage void call_debug(void);
  75
  76 struct notifier_block *die_chain;
  77 static DEFINE_SPINLOCK(die_notifier_lock);
  78
  79 int register_die_notifier(struct notifier_block *nb)
  80 {
  81         int err = 0;
  82         unsigned long flags;
  83         spin_lock_irqsave(&die_notifier_lock, flags);
  84         err = notifier_chain_register(&die_chain, nb);
  85         spin_unlock_irqrestore(&die_notifier_lock, flags);
  86         return err;
  87 }
  88
  89 static inline void conditional_sti(struct pt_regs *regs)
  90 {
  91         if (regs->eflags & X86_EFLAGS_IF)
  92                 local_irq_enable();
  93 }
  94
  95 static int kstack_depth_to_print = 10;
  96
  97 #ifdef CONFIG_KALLSYMS
  98 #include <linux/kallsyms.h>
  99 int printk_address(unsigned long address)
 100 {
 101         unsigned long offset = 0, symsize;
 102         const char *symname;
 103         char *modname;
 104         char *delim = ":";
 105         char namebuf[128];
 106
 107         symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
 108         if (!symname)
 109                 return printk("[<%016lx>]", address);
 110         if (!modname)
 111                 modname = delim = "";
 112         return printk("<%016lx>{%s%s%s%s%+ld}",
 113                       address,delim,modname,delim,symname,offset);
 114 }
 115 #else
 116 int printk_address(unsigned long address)
 117 {
 118         return printk("[<%016lx>]", address);
 119 }
 120 #endif
 121
 122 unsigned long *in_exception_stack(int cpu, unsigned long stack)
 123 {
 124         int k;
 125         for (k = 0; k < N_EXCEPTION_STACKS; k++) {
 126                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 127                 unsigned long start = tss->ist[k] - EXCEPTION_STKSZ;
 128
 129                 if (stack >= start && stack < tss->ist[k])
 130                         return (unsigned long *)tss->ist[k];
 131         }
 132         return NULL;
 133 }
 134
 135 /*
 136  * x86-64 can have upto three kernel stacks:
 137  * process stack
 138  * interrupt stack
 139  * severe exception (double fault, nmi, stack fault) hardware stack
 140  * Check and process them in order.
 141  */
 142
 143 void show_trace(unsigned long *stack)
 144 {
 145         unsigned long addr;
 146         unsigned long *irqstack, *irqstack_end, *estack_end;
 147         const int cpu = safe_smp_processor_id();
 148         int i;
 149
 150         printk("\nCall Trace:");
 151         i = 0;
 152
 153         estack_end = in_exception_stack(cpu, (unsigned long)stack);
 154         if (estack_end) {
 155                 while (stack < estack_end) {
 156                         addr = *stack++;
 157                         if (__kernel_text_address(addr)) {
 158                                 i += printk_address(addr);
 159                                 i += printk(" ");
 160                                 if (i > 50) {
 161                                         printk("\n");
 162                                         i = 0;
 163                                 }
 164                         }
 165                 }
 166                 i += printk(" <EOE> ");
 167                 i += 7;
 168                 stack = (unsigned long *) estack_end[-2];
 169         }
 170
 171         irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 172         irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE + 64);
 173
 174         if (stack >= irqstack && stack < irqstack_end) {
 175                 printk("<IRQ> ");
 176                 while (stack < irqstack_end) {
 177                         addr = *stack++;
 178                         /*
 179                          * If the address is either in the text segment of the
 180                          * kernel, or in the region which contains vmalloc'ed
 181                          * memory, it *may* be the address of a calling
 182                          * routine; if so, print it so that someone tracing
 183                          * down the cause of the crash will be able to figure
 184                          * out the call path that was taken.
 185                          */
 186                          if (__kernel_text_address(addr)) {
 187                                  i += printk_address(addr);
 188                                  i += printk(" ");
 189                                  if (i > 50) {
 190                                         printk("\n       ");
 191                                          i = 0;
 192                                  }
 193                         }
 194                 }
 195                 stack = (unsigned long *) (irqstack_end[-1]);
 196                 printk(" <EOI> ");
 197                 i += 7;
 198         }
 199
 200         while (((long) stack & (THREAD_SIZE-1)) != 0) {
 201                 addr = *stack++;
 202                 if (__kernel_text_address(addr)) {
 203                         i += printk_address(addr);
 204                         i += printk(" ");
 205                         if (i > 50) {
 206                                 printk("\n       ");
 207                                          i = 0;
 208                         }
 209                 }
 210         }
 211         printk("\n");
 212 }
 213
 214 void show_stack(struct task_struct *tsk, unsigned long * rsp)
 215 {
 216         unsigned long *stack;
 217         int i;
 218         const int cpu = safe_smp_processor_id();
 219         unsigned long *irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr);
 220         unsigned long *irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE);
 221
 222         // debugging aid: "show_stack(NULL, NULL);" prints the
 223         // back trace for this cpu.
 224
 225         if (rsp == NULL) {
 226                 if (tsk)
 227                         rsp = (unsigned long *)tsk->thread.rsp;
 228                 else
 229                         rsp = (unsigned long *)&rsp;
 230         }
 231
 232         stack = rsp;
 233         for(i=0; i < kstack_depth_to_print; i++) {
 234                 if (stack >= irqstack && stack <= irqstack_end) {
 235                         if (stack == irqstack_end) {
 236                                 stack = (unsigned long *) (irqstack_end[-1]);
 237                                 printk(" <EOI> ");
 238                         }
 239                 } else {
 240                 if (((long) stack & (THREAD_SIZE-1)) == 0)
 241                         break;
 242                 }
 243                 if (i && ((i % 4) == 0))
 244                         printk("\n       ");
 245                 printk("%016lx ", *stack++);
 246         }
 247         show_trace((unsigned long *)rsp);
 248 }
 249
 250 /*
 251  * The architecture-independent dump_stack generator
 252  */
 253 void dump_stack(void)
 254 {
 255         unsigned long dummy;
 256         show_trace(&dummy);
 257 }
 258
 259 EXPORT_SYMBOL(dump_stack);
 260
 261 void show_registers(struct pt_regs *regs)
 262 {
 263         int i;
 264         int in_kernel = (regs->cs & 3) == 0;
 265         unsigned long rsp;
 266         const int cpu = safe_smp_processor_id();
 267         struct task_struct *cur = cpu_pda[cpu].pcurrent;
 268
 269                 rsp = regs->rsp;
 270
 271         printk("CPU %d ", cpu);
 272         __show_regs(regs);
 273         printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
 274                 cur->comm, cur->pid, cur->thread_info, cur);
 275
 276         /*
 277          * When in-kernel, we also print out the stack and code at the
 278          * time of the fault..
 279          */
 280         if (in_kernel) {
 281
 282                 printk("Stack: ");
 283                 show_stack(NULL, (unsigned long*)rsp);
 284
 285                 printk("\nCode: ");
 286                 if(regs->rip < PAGE_OFFSET)
 287                         goto bad;
 288
 289                 for(i=0;i<20;i++)
 290                 {
 291                         unsigned char c;
 292                         if(__get_user(c, &((unsigned char*)regs->rip)[i])) {
 293 bad:
 294                                 printk(" Bad RIP value.");
 295                                 break;
 296                         }
 297                         printk("%02x ", c);
 298                 }
 299         }
 300         printk("\n");
 301 }
 302
 303 void handle_BUG(struct pt_regs *regs)
 304 {
 305         struct bug_frame f;
 306         char tmp;
 307
 308         if (regs->cs & 3)
 309                 return;
 310         if (__copy_from_user(&f, (struct bug_frame *) regs->rip,
 311                              sizeof(struct bug_frame)))
 312                 return;
 313         if ((unsigned long)f.filename < __PAGE_OFFSET ||
 314             f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
 315                 return;
 316         if (__get_user(tmp, f.filename))
 317                 f.filename = "unmapped filename";
 318         printk("----------- [cut here ] --------- [please bite here ] ---------\n");
 319         printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", f.filename, f.line);
 320 }
 321
 322 void out_of_line_bug(void)
 323 {
 324         BUG();
 325 }
 326
 327 static DEFINE_SPINLOCK(die_lock);
 328 static int die_owner = -1;
 329
 330 void oops_begin(void)
 331 {
 332         int cpu = safe_smp_processor_id();
 333         /* racy, but better than risking deadlock. */
 334         local_irq_disable();
 335         if (!spin_trylock(&die_lock)) {
 336                 if (cpu == die_owner)
 337                         /* nested oops. should stop eventually */;
 338                 else
 339                         spin_lock(&die_lock);
 340         }
 341         die_owner = cpu;
 342         console_verbose();
 343         bust_spinlocks(1);
 344 }
 345
 346 void oops_end(void)
 347 {
 348         die_owner = -1;
 349         bust_spinlocks(0);
 350         spin_unlock(&die_lock);
 351         if (panic_on_oops)
 352                 panic("Oops");
 353 }
 354
 355 void __die(const char * str, struct pt_regs * regs, long err)
 356 {
 357         static int die_counter;
 358         printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
 359 #ifdef CONFIG_PREEMPT
 360         printk("PREEMPT ");
 361 #endif
 362 #ifdef CONFIG_SMP
 363         printk("SMP ");
 364 #endif
 365 #ifdef CONFIG_DEBUG_PAGEALLOC
 366         printk("DEBUG_PAGEALLOC");
 367 #endif
 368         printk("\n");
 369         notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
 370         show_registers(regs);
 371         /* Executive summary in case the oops scrolled away */
 372         printk(KERN_ALERT "RIP ");
 373         printk_address(regs->rip);
 374         printk(" RSP <%016lx>\n", regs->rsp);
 375 }
 376
 377 void die(const char * str, struct pt_regs * regs, long err)
 378 {
 379         oops_begin();
 380         handle_BUG(regs);
 381         __die(str, regs, err);
 382         oops_end();
 383         do_exit(SIGSEGV);
 384 }
 385 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
 386 {
 387         if (!(regs->eflags & VM_MASK) && (regs->cs == __KERNEL_CS))
 388                 die(str, regs, err);
 389 }
 390
 391 void die_nmi(char *str, struct pt_regs *regs)
 392 {
 393         oops_begin();
 394         /*
 395          * We are in trouble anyway, lets at least try
 396          * to get a message out.
 397          */
 398         printk(str, safe_smp_processor_id());
 399         show_registers(regs);
 400         if (panic_on_timeout || panic_on_oops)
 401                 panic("nmi watchdog");
 402         printk("console shuts up ...\n");
 403         oops_end();
 404         do_exit(SIGSEGV);
 405 }
 406
 407 static void do_trap(int trapnr, int signr, char *str,
 408                            struct pt_regs * regs, long error_code, siginfo_t *info)
 409 {
 410         conditional_sti(regs);
 411
 412 #ifdef CONFIG_CHECKING
 413        {
 414                unsigned long gs;
 415                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 416                rdmsrl(MSR_GS_BASE, gs);
 417                if (gs != (unsigned long)pda) {
 418                        wrmsrl(MSR_GS_BASE, pda);
 419                        printk("%s: wrong gs %lx expected %p rip %lx\n", str, gs, pda,
 420                               regs->rip);
 421                }
 422        }
 423 #endif
 424
 425         if ((regs->cs & 3)  != 0) {
 426                 struct task_struct *tsk = current;
 427
 428                 if (exception_trace && unhandled_signal(tsk, signr))
 429                         printk(KERN_INFO
 430                                "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
 431                                tsk->comm, tsk->pid, str,
 432                                regs->rip,regs->rsp,error_code);
 433
 434                 tsk->thread.error_code = error_code;
 435                 tsk->thread.trap_no = trapnr;
 436                 if (info)
 437                         force_sig_info(signr, info, tsk);
 438                 else
 439                         force_sig(signr, tsk);
 440                 return;
 441         }
 442
 443
 444         /* kernel trap */
 445         {
 446                 const struct exception_table_entry *fixup;
 447                 fixup = search_exception_tables(regs->rip);
 448                 if (fixup) {
 449                         regs->rip = fixup->fixup;
 450                 } else
 451                         die(str, regs, error_code);
 452                 return;
 453         }
 454 }
 455
 456 #define DO_ERROR(trapnr, signr, str, name) \
 457 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 458 { \
 459         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 460                                                         == NOTIFY_STOP) \
 461                 return; \
 462         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 463 }
 464
 465 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
 466 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
 467 { \
 468         siginfo_t info; \
 469         info.si_signo = signr; \
 470         info.si_errno = 0; \
 471         info.si_code = sicode; \
 472         info.si_addr = (void __user *)siaddr; \
 473         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 474                                                         == NOTIFY_STOP) \
 475                 return; \
 476         do_trap(trapnr, signr, str, regs, error_code, &info); \
 477 }
 478
 479 DO_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->rip)
 480 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
 481 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
 482 DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->rip)
 483 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
 484 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
 485 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 486 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
 487 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 488 DO_ERROR(18, SIGSEGV, "reserved", reserved)
 489
 490 #define DO_ERROR_STACK(trapnr, signr, str, name) \
 491 asmlinkage void *do_##name(struct pt_regs * regs, long error_code) \
 492 { \
 493         struct pt_regs *pr = ((struct pt_regs *)(current->thread.rsp0))-1; \
 494         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
 495                                                         == NOTIFY_STOP) \
 496                 return regs; \
 497         if (regs->cs & 3) { \
 498                 memcpy(pr, regs, sizeof(struct pt_regs)); \
 499                 regs = pr; \
 500         } \
 501         do_trap(trapnr, signr, str, regs, error_code, NULL); \
 502         return regs;            \
 503 }
 504
 505 DO_ERROR_STACK(12, SIGBUS,  "stack segment", stack_segment)
 506 DO_ERROR_STACK( 8, SIGSEGV, "double fault", double_fault)
 507
 508 asmlinkage void do_general_protection(struct pt_regs * regs, long error_code)
 509 {
 510         conditional_sti(regs);
 511
 512 #ifdef CONFIG_CHECKING
 513        {
 514                unsigned long gs;
 515                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 516                rdmsrl(MSR_GS_BASE, gs);
 517                if (gs != (unsigned long)pda) {
 518                        wrmsrl(MSR_GS_BASE, pda);
 519                        oops_in_progress++;
 520                        printk("general protection handler: wrong gs %lx expected %p\n", gs, pda);
 521                        oops_in_progress--;
 522                }
 523        }
 524 #endif
 525
 526         if ((regs->cs & 3)!=0) {
 527                 struct task_struct *tsk = current;
 528
 529                 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
 530                         printk(KERN_INFO
 531                        "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
 532                                tsk->comm, tsk->pid,
 533                                regs->rip,regs->rsp,error_code);
 534
 535                 tsk->thread.error_code = error_code;
 536                 tsk->thread.trap_no = 13;
 537                 force_sig(SIGSEGV, tsk);
 538                 return;
 539         }
 540
 541         /* kernel gp */
 542         {
 543                 const struct exception_table_entry *fixup;
 544                 fixup = search_exception_tables(regs->rip);
 545                 if (fixup) {
 546                         regs->rip = fixup->fixup;
 547                         return;
 548                 }
 549                 if (notify_die(DIE_GPF, "general protection fault", regs,
 550                                         error_code, 13, SIGSEGV) == NOTIFY_STOP)
 551                         return;
 552                 die("general protection fault", regs, error_code);
 553         }
 554 }
 555
 556 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
 557 {
 558         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
 559         printk("You probably have a hardware problem with your RAM chips\n");
 560
 561         /* Clear and disable the memory parity error line. */
 562         reason = (reason & 0xf) | 4;
 563         outb(reason, 0x61);
 564 }
 565
 566 static void io_check_error(unsigned char reason, struct pt_regs * regs)
 567 {
 568         printk("NMI: IOCK error (debug interrupt?)\n");
 569         show_registers(regs);
 570
 571         /* Re-enable the IOCK line, wait for a few seconds */
 572         reason = (reason & 0xf) | 8;
 573         outb(reason, 0x61);
 574         mdelay(2000);
 575         reason &= ~8;
 576         outb(reason, 0x61);
 577 }
 578
 579 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
 580 {       printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
 581         printk("Dazed and confused, but trying to continue\n");
 582         printk("Do you have a strange power saving mode enabled?\n");
 583 }
 584
 585 asmlinkage void default_do_nmi(struct pt_regs *regs)
 586 {
 587         unsigned char reason = 0;
 588
 589         /* Only the BSP gets external NMIs from the system.  */
 590         if (!smp_processor_id())
 591                 reason = get_nmi_reason();
 592
 593         if (!(reason & 0xc0)) {
 594                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
 595                                                                 == NOTIFY_STOP)
 596                         return;
 597 #ifdef CONFIG_X86_LOCAL_APIC
 598                 /*
 599                  * Ok, so this is none of the documented NMI sources,
 600                  * so it must be the NMI watchdog.
 601                  */
 602                 if (nmi_watchdog > 0) {
 603                         nmi_watchdog_tick(regs,reason);
 604                         return;
 605                 }
 606 #endif
 607                 unknown_nmi_error(reason, regs);
 608                 return;
 609         }
 610         if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
 611                 return;
 612
 613         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 614
 615         if (reason & 0x80)
 616                 mem_parity_error(reason, regs);
 617         if (reason & 0x40)
 618                 io_check_error(reason, regs);
 619 }
 620
 621 asmlinkage void do_int3(struct pt_regs * regs, long error_code)
 622 {
 623         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
 624                 return;
 625         }
 626         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 627         return;
 628 }
 629
 630 /* runs on IST stack. */
 631 asmlinkage void *do_debug(struct pt_regs * regs, unsigned long error_code)
 632 {
 633         struct pt_regs *pr;
 634         unsigned long condition;
 635         struct task_struct *tsk = current;
 636         siginfo_t info;
 637
 638         pr = (struct pt_regs *)(current->thread.rsp0)-1;
 639         if (regs->cs & 3) {
 640                 memcpy(pr, regs, sizeof(struct pt_regs));
 641                 regs = pr;
 642         }
 643
 644 #ifdef CONFIG_CHECKING
 645        {
 646                /* RED-PEN interaction with debugger - could destroy gs */
 647                unsigned long gs;
 648                struct x8664_pda *pda = cpu_pda + safe_smp_processor_id();
 649                rdmsrl(MSR_GS_BASE, gs);
 650                if (gs != (unsigned long)pda) {
 651                        wrmsrl(MSR_GS_BASE, pda);
 652                        printk("debug handler: wrong gs %lx expected %p\n", gs, pda);
 653                }
 654        }
 655 #endif
 656
 657         asm("movq %%db6,%0" : "=r" (condition));
 658
 659         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 660                                                 SIGTRAP) == NOTIFY_STOP) {
 661                 return regs;
 662         }
 663         conditional_sti(regs);
 664
 665         /* Mask out spurious debug traps due to lazy DR7 setting */
 666         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 667                 if (!tsk->thread.debugreg7) {
 668                         goto clear_dr7;
 669                 }
 670         }
 671
 672         tsk->thread.debugreg6 = condition;
 673
 674         /* Mask out spurious TF errors due to lazy TF clearing */
 675         if ((condition & DR_STEP) &&
 676             (notify_die(DIE_DEBUGSTEP, "debugstep", regs, condition,
 677                         1, SIGTRAP) != NOTIFY_STOP)) {
 678                 /*
 679                  * The TF error should be masked out only if the current
 680                  * process is not traced and if the TRAP flag has been set
 681                  * previously by a tracing process (condition detected by
 682                  * the PT_DTRACE flag); remember that the i386 TRAP flag
 683                  * can be modified by the process itself in user mode,
 684                  * allowing programs to debug themselves without the ptrace()
 685                  * interface.
 686                  */
 687                 if ((regs->cs & 3) == 0)
 688                        goto clear_TF_reenable;
 689                 if ((tsk->ptrace & (PT_DTRACE|PT_PTRACED)) == PT_DTRACE)
 690                         goto clear_TF;
 691         }
 692
 693         /* Ok, finally something we can handle */
 694         tsk->thread.trap_no = 1;
 695         tsk->thread.error_code = error_code;
 696         info.si_signo = SIGTRAP;
 697         info.si_errno = 0;
 698         info.si_code = TRAP_BRKPT;
 699         if ((regs->cs & 3) == 0)
 700                 goto clear_dr7;
 701
 702         info.si_addr = (void __user *)regs->rip;
 703         force_sig_info(SIGTRAP, &info, tsk);
 704 clear_dr7:
 705         asm volatile("movq %0,%%db7"::"r"(0UL));
 706         notify_die(DIE_DEBUG, "debug", regs, condition, 1, SIGTRAP);
 707         return regs;
 708
 709 clear_TF_reenable:
 710         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 711
 712 clear_TF:
 713         /* RED-PEN could cause spurious errors */
 714         if (notify_die(DIE_DEBUG, "debug2", regs, condition, 1, SIGTRAP)
 715                                                                 != NOTIFY_STOP)
 716         regs->eflags &= ~TF_MASK;
 717         return regs;
 718 }
 719
 720 static int kernel_math_error(struct pt_regs *regs, char *str)
 721 {
 722         const struct exception_table_entry *fixup;
 723         fixup = search_exception_tables(regs->rip);
 724         if (fixup) {
 725                 regs->rip = fixup->fixup;
 726                 return 1;
 727         }
 728         notify_die(DIE_GPF, str, regs, 0, 16, SIGFPE);
 729 #if 0
 730         /* This should be a die, but warn only for now */
 731         die(str, regs, 0);
 732 #else
 733         printk(KERN_DEBUG "%s: %s at ", current->comm, str);
 734         printk_address(regs->rip);
 735         printk("\n");
 736 #endif
 737         return 0;
 738 }
 739
 740 /*
 741  * Note that we play around with the 'TS' bit in an attempt to get
 742  * the correct behaviour even in the presence of the asynchronous
 743  * IRQ13 behaviour
 744  */
 745 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
 746 {
 747         void __user *rip = (void __user *)(regs->rip);
 748         struct task_struct * task;
 749         siginfo_t info;
 750         unsigned short cwd, swd;
 751
 752         conditional_sti(regs);
 753         if ((regs->cs & 3) == 0 &&
 754             kernel_math_error(regs, "kernel x87 math error"))
 755                 return;
 756
 757         /*
 758          * Save the info for the exception handler and clear the error.
 759          */
 760         task = current;
 761         save_init_fpu(task);
 762         task->thread.trap_no = 16;
 763         task->thread.error_code = 0;
 764         info.si_signo = SIGFPE;
 765         info.si_errno = 0;
 766         info.si_code = __SI_FAULT;
 767         info.si_addr = rip;
 768         /*
 769          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 770          * status.  0x3f is the exception bits in these regs, 0x200 is the
 771          * C1 reg you need in case of a stack fault, 0x040 is the stack
 772          * fault bit.  We should only be taking one exception at a time,
 773          * so if this combination doesn't produce any single exception,
 774          * then we have a bad program that isn't synchronizing its FPU usage
 775          * and it will suffer the consequences since we won't be able to
 776          * fully reproduce the context of the exception
 777          */
 778         cwd = get_fpu_cwd(task);
 779         swd = get_fpu_swd(task);
 780         switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
 781                 case 0x000:
 782                 default:
 783                         break;
 784                 case 0x001: /* Invalid Op */
 785                 case 0x041: /* Stack Fault */
 786                 case 0x241: /* Stack Fault | Direction */
 787                         info.si_code = FPE_FLTINV;
 788                         break;
 789                 case 0x002: /* Denormalize */
 790                 case 0x010: /* Underflow */
 791                         info.si_code = FPE_FLTUND;
 792                         break;
 793                 case 0x004: /* Zero Divide */
 794                         info.si_code = FPE_FLTDIV;
 795                         break;
 796                 case 0x008: /* Overflow */
 797                         info.si_code = FPE_FLTOVF;
 798                         break;
 799                 case 0x020: /* Precision */
 800                         info.si_code = FPE_FLTRES;
 801                         break;
 802         }
 803         force_sig_info(SIGFPE, &info, task);
 804 }
 805
 806 asmlinkage void bad_intr(void)
 807 {
 808         printk("bad interrupt");
 809 }
 810
 811 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
 812 {
 813         void __user *rip = (void __user *)(regs->rip);
 814         struct task_struct * task;
 815         siginfo_t info;
 816         unsigned short mxcsr;
 817
 818         conditional_sti(regs);
 819         if ((regs->cs & 3) == 0 &&
 820                 kernel_math_error(regs, "simd math error"))
 821                 return;
 822
 823         /*
 824          * Save the info for the exception handler and clear the error.
 825          */
 826         task = current;
 827         save_init_fpu(task);
 828         task->thread.trap_no = 19;
 829         task->thread.error_code = 0;
 830         info.si_signo = SIGFPE;
 831         info.si_errno = 0;
 832         info.si_code = __SI_FAULT;
 833         info.si_addr = rip;
 834         /*
 835          * The SIMD FPU exceptions are handled a little differently, as there
 836          * is only a single status/control register.  Thus, to determine which
 837          * unmasked exception was caught we must mask the exception mask bits
 838          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 839          */
 840         mxcsr = get_fpu_mxcsr(task);
 841         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 842                 case 0x000:
 843                 default:
 844                         break;
 845                 case 0x001: /* Invalid Op */
 846                         info.si_code = FPE_FLTINV;
 847                         break;
 848                 case 0x002: /* Denormalize */
 849                 case 0x010: /* Underflow */
 850                         info.si_code = FPE_FLTUND;
 851                         break;
 852                 case 0x004: /* Zero Divide */
 853                         info.si_code = FPE_FLTDIV;
 854                         break;
 855                 case 0x008: /* Overflow */
 856                         info.si_code = FPE_FLTOVF;
 857                         break;
 858                 case 0x020: /* Precision */
 859                         info.si_code = FPE_FLTRES;
 860                         break;
 861         }
 862         force_sig_info(SIGFPE, &info, task);
 863 }
 864
 865 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
 866 {
 867 }
 868
 869 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 870 {
 871 }
 872
 873 /*
 874  *  'math_state_restore()' saves the current math information in the
 875  * old math state array, and gets the new ones from the current task
 876  *
 877  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 878  * Don't touch unless you *really* know how it works.
 879  */
 880 asmlinkage void math_state_restore(void)
 881 {
 882         struct task_struct *me = current;
 883         clts();                 /* Allow maths ops (or we recurse) */
 884
 885         if (!used_math())
 886                 init_fpu(me);
 887         restore_fpu_checking(&me->thread.i387.fxsave);
 888         me->thread_info->status |= TS_USEDFPU;
 889 }
 890
 891 void do_call_debug(struct pt_regs *regs)
 892 {
 893         notify_die(DIE_CALL, "debug call", regs, 0, 255, SIGINT);
 894 }
 895
 896 void __init trap_init(void)
 897 {
 898         set_intr_gate(0,&divide_error);
 899         set_intr_gate_ist(1,&debug,DEBUG_STACK);
 900         set_intr_gate_ist(2,&nmi,NMI_STACK);
 901         set_system_gate(3,&int3);
 902         set_system_gate(4,&overflow);   /* int4-5 can be called from all */
 903         set_system_gate(5,&bounds);
 904         set_intr_gate(6,&invalid_op);
 905         set_intr_gate(7,&device_not_available);
 906         set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
 907         set_intr_gate(9,&coprocessor_segment_overrun);
 908         set_intr_gate(10,&invalid_TSS);
 909         set_intr_gate(11,&segment_not_present);
 910         set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
 911         set_intr_gate(13,&general_protection);
 912         set_intr_gate(14,&page_fault);
 913         set_intr_gate(15,&spurious_interrupt_bug);
 914         set_intr_gate(16,&coprocessor_error);
 915         set_intr_gate(17,&alignment_check);
 916 #ifdef CONFIG_X86_MCE
 917         set_intr_gate_ist(18,&machine_check, MCE_STACK);
 918 #endif
 919         set_intr_gate(19,&simd_coprocessor_error);
 920
 921 #ifdef CONFIG_IA32_EMULATION
 922         set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 923 #endif
 924
 925         set_intr_gate(KDB_VECTOR, call_debug);
 926
 927         /*
 928          * Should be a barrier for any external CPU state.
 929          */
 930         cpu_init();
 931 }
 932
 933
 934 /* Actual parsing is done early in setup.c. */
 935 static int __init oops_dummy(char *s)
 936 {
 937         panic_on_oops = 1;
 938         return -1;
 939 }
 940 __setup("oops=", oops_dummy);
 941
 942 static int __init kstack_setup(char *s)
 943 {
 944         kstack_depth_to_print = simple_strtoul(s,NULL,0);
 945         return 0;
 946 }
 947 __setup("kstack=", kstack_setup);
 948